title: “Violence at the Oscars”

author: “Otobo Ujah”

date: “4/24/2022”

output:

html_document: default

Project Summary

The slap at the Oscars by Will Smith has generated mixed reactions from the public. This project explores sentiments of the public towards violence at the 94th Academy awards using data from twitter

Research Question

Following the 94th Academy Awards, what and how are the sentiments and underlying themes of the public as expressed on Twitter?

Import packages for analysis

library(tidytext)
library(magrittr)
library(plotly)
library(tidyverse)
library(ggthemes)
library(rtweet) 
library(syuzhet)
library(see)
library(here)
library(tidytext)
library(tm)
library(wordcloud)
library(reshape2)
library(ggplot2)
library(see)
library(SnowballC)
library(wordcloud)

Import data

osc <- read_csv(here::here("data", "oscars.csv"))

## New names:
## * `` -> ...1

## Rows: 450207 Columns: 2

## -- Column specification --------------------------------------------------------
## Delimiter: ","
## chr (1): text
## dbl (1): ...1

## 
## i Use `spec()` to retrieve the full column specification for this data.
## i Specify the column types or set `show_col_types = FALSE` to quiet this message.

Clean the data

osc$text = gsub("@\\w+", "", osc$text)

osc$text = gsub('@\\S+', '', osc$text)

osc$text<- gsub("[[:digit:]]", "", osc$text) 

osc$text <- gsub("^[[:space:]]*","",osc$text) ## Remove leading whitespaces

osc$text <- gsub("[[:space:]]*$","",osc$text) ## Remove trailing whitespaces

osc$text <- gsub(' +',' ',osc$text) 

oscars <- 
  osc |> 
  select(text) |>
  mutate(text = text |>  str_to_lower())|>
  mutate(text = text |>  str_remove_all(pattern = '\\n')) |>
  mutate(text = text |>  str_remove_all(pattern = '&amp')) |>
  mutate(text = text |>  str_remove_all(pattern = 'https://t.co/[a-z,A-Z,0-9]*')) |>
  mutate(text = text |>  str_remove_all(pattern = 'https')) |>
  mutate(text = text |>  str_remove(pattern = '^(rt)')) |>
  mutate(text = text |>  str_remove_all(pattern = '\\_'))|>
  mutate(text = text |>  str_remove_all(pattern = '[:emoji:]'))


oscars$text <- gsub("[[:punct:]]", "", oscars$text)

oscars$text <- gsub("[^[:alnum:]]", " ", oscars$text)

oscars$text <- gsub(' +',' ',oscars$text) ## Remove extra whitespaces

oscars["DuplicateFlag"] = duplicated(oscars$text)
oscars = subset(oscars, oscars$DuplicateFlag=="FALSE")
oscars = subset(oscars, select = -c(DuplicateFlag))   


oscars <- 
  oscars|> 
  na_if("")|>
  na.omit()

oscars<- oscars[!duplicated(gsub("^(\\S+\\s+\\S+\\s+\\S+).*", "\\1", oscars$text)),]

Unnesting and Removing Stopwords

sw <- c("rt","íí","get","didn","like","just","yes","know","will","good","day","people","doesn’t", "s","you’re","i’ll","he’s","doesn’t","dont","don't", "didnt", "wasnt" ,"yall","isnt", "doesnt", "hes", "wouldnt", "didn’t","it’s","wasn’t","wasn", "wouldn","zz","zzz","zzzz", "zzzzz", "zzzzzz", "zzzzzzz", "zzzzzzzz", "zzzzzzzzzz", "zyx", "wouldn’t", "abc","i’m", "youre", "im")

stopwords.df <- tibble(
  word = c(stopwords(kind = 'en'),  
           sw))

colnames(stopwords.df) <- "word"

words <- 
  oscars|>
  select(text)|>
  unnest_tokens(input = text, output = word) |>
  anti_join(y = stopwords.df, by = 'word')|>
  anti_join(stop_words) |>
  filter(!word %in% stop_words$word)

## Joining, by = "word"

word_count <- words |>
  count(word, sort = TRUE) |>
  filter(n>1)

Plot of word count

plot <- word_count |>
  filter(word!="oscars")|>
  filter(word!="chris")|>
  filter(word!="rock")|>
  filter(word!="willsmith")|>
  filter(word!="oscar")|>
  filter(word!="jada")|>
  filter(word!="chrisrock")|>
  filter(word!="smith")|>
  filter(word!="oscars")|>
  filter(word!="academyawards")|>
  filter(word!="awards")|>
  filter(word!="academy")|>
  filter(word!="willandchris") |>
  top_n(30) |>
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(x = word, y = (n/(sum(n)))*100, fill = word)) +
  geom_bar(stat = 'identity') +
  ylab("Count (%)") +
  xlab(NULL)+
  labs(title = 'Fig. 1: Most common words in tweets') +
  coord_flip()+
  theme_pander() +
  theme(legend.position="none",
        axis.text.y = element_text(color = "grey20", 
                                   size = 20, 
                                   angle = 0, 
                                   hjust = 1, 
                                   vjust = 0, 
                                   face = "plain"),
        plot.title = element_text(size=22)) 


plot |> ggplotly()

set.seed(222)

wc <- word_count |>
  # Set count threshold. 
  filter(word!="oscars")|>
  filter(word!="chris")|>
  filter(word!="rock")|>
  filter(word!="willsmith")|>
  filter(word!="oscar")|>
  filter(word!="jada")|>
  filter(word!="chrisrock")|>
  filter(word!="smith")|>
  filter(word!="oscars")|>
  filter(word!="academyawards")|>
  filter(word!="academyaward")|>
  filter(word!="awards")|>
  filter(word!="academy")|>
  filter(word!="jadapinkettsmith")|>
  filter(word!="willandchris")|>
  filter(word!="ive")|>
  filter(word!="isn")|>
  filter(word!="aint")|>
  filter(word!="wont")|>
  filter(word!="ill")|>
  filter(word!="ain")|>
  filter(word!="ve")|>
  filter(word!="whats")|>
  filter(word!="id")

set.seed(123)
wordcloud(words = wc$word, 
          freq = wc$n,
          min.freq = 1500,
          random.order=FALSE,
          random.color = TRUE,
          rot.per=0.3, 
          scale= c(4,1), 
          colors = brewer.pal(7, 'Dark2'))

SENTIMENT ANALYSIS USING THE BING LEXICON

bing <- get_sentiments("bing")

bing.df <- oscars %>% 
  unnest_tokens(input = text, output = word) %>% 
  anti_join(y = stopwords.df, by = 'word')|>
  anti_join(stop_words) |>
  filter(!word %in% stop_words$word)|>
  filter(!word %in% stopwords.df$word)|>
  inner_join(bing)

## Joining, by = "word"
## Joining, by = "word"

checks <- bing.df |>
  mutate(word = SnowballC::wordStem(word))

class(checks$word)

## [1] "character"

colnames(checks) <- c( "word", "sentiment")

check <- checks |>
  inner_join(bing) |>
  select(sentiment) |>
  count(sentiment)|>
  mutate(sentiment = reorder(sentiment, n))|>
  ggplot(aes(x = sentiment, y = (n/sum(n))*100, fill = sentiment)) +
  coord_flip()+ 
  geom_bar(stat = 'identity')  +
  xlab("Words") +
  ylab("Count (%)") +
  theme(text = element_text(size = 20),
        axis.text.y = element_text(color = "grey20",
                                   size = 20, 
                                   angle = 0,
                                   hjust = 1,
                                   vjust = 0,
                                   face = "plain")) +
  ggtitle(label = 'Fig. 3: Overall sentiments of tweets using the Bing et al. Lexicon')

## Joining, by = c("word", "sentiment")

check

bingcounts <-oscars %>% 
  unnest_tokens(input = text, output = word) %>% 
  anti_join(y = stopwords.df, by = 'word')|>
  anti_join(stop_words) |>
  filter(!word %in% stop_words$word)|>
  filter(!word %in% stopwords.df$word)|>
  inner_join(bing)|>
  mutate(word = SnowballC::wordStem(word))|>
  count(word, sentiment, sort = TRUE) %>%
  ungroup()

## Joining, by = "word"
## Joining, by = "word"

bing_sent <- 
  bingcounts %>%
  group_by(sentiment) %>%
  top_n(20) %>%
  ungroup() %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(word, y = (n/sum(n))*100, fill = sentiment)) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~sentiment, scales = "free_y") +
  coord_flip() +
  xlab("Words") +
  ylab("Count (%)") +
  ggtitle( "Fig. 4:Top 20 word contribution per sentiment")+
  theme(text = element_text(size = 20)) 

## Selecting by n

bing_sent

bing2 <- 
  bingcounts |>
  top_n(30) |>
  mutate(n =  (n/sum(n))*100)|>
  mutate(n = ifelse(sentiment=="negative", -n,n))|>
  ggplot(aes(fct_reorder(str_to_title(word),n),n, fill = str_to_title(sentiment))) +
  geom_col() +
  coord_flip() +
  scale_fill_brewer(type = "qual") +
  guides(fill = guide_legend(reverse = T)) +
  labs(title = "Fig. 5: Classification of top 30 words into positive and negative sentiments based on Bing et. al lexicon",
       y = "Compund sentiment score (%)",
       x = "",
       fill = "Sentiment") +
  theme(text = element_text(size = 20))

## Selecting by n

ggsave(here::here("bbing2.svg"), 
       bing2,
       height = 6,
       width = 6)

bing2

Results

From March 28 to March 30, 2020, 715,164 tweets related to violence at the 2022 Oscar awards were retrieved. After preprocessing, the final corpus consisted of 305769 tweets corresponding to a total of 2782085 words. Figure 1 shows the words which occurred most frequently. Words and phrases included in the search terms were excluded given their frequency of occurrence in the tweets and, which consequently have greater weightings relative to other words in the corpus of tweets. The word frequencies from the corpus of tweets about violence during the 2022 Oscar awards ceremony were analyzed using a word cloud as shown in the figure. Word clouds enable visualization of frequently occurring words based on their level of prominence within a corpus of text, with the sizes of words reflecting the frequency of its occurrence in the corpus. The words “slap”, “joke”, “night”, “watch” , “wifeviolence”, “black” and, “stage” was among the most frequently occurring words. Bigrams were also created to determine the connections between words. After tokenizing words, those with low count were considered to be either misspelled, obscure or non-words and hence non-informative. These were removed by filtering out words with frequencies, n < 10. Based on the Bing et al lexicon, 63.2% of the overall sentiments were expressed as negative while 36.7% were positive sentiments. The top 20 words contributing to each sentiment category was further explored. Among the negative sentiments expressed, the words “slap”, “joke”, “shit”, “assault” and, “bad” were among the frequently occurring words while “won”, “award”, “love”, “awards” and “win” were among the frequently occurring words among the positive sentiments expressed. These findings are further depicted in figures 4 and 5.